At present, according to different scales, a resource management server system generally includes tens, hundreds or even tens of thousands of computer server nodes. Most of these server nodes are commercially available servers up to commercial standards, and have independent hardware devices such as memory.
With the increasing popularity and growing acceptance of a cloud computing model, a server node needs to serve cloud computing, and therefore, application loads become increasingly varied. Compared with conventional loads such as transaction processing and network services, newly emerged workloads are all data-centric. These new input/output (IO)-intensive loads are run on a large-scale cluster (including hundreds or thousands of servers), and varied (structured, unstructured, and multimedia) data is processed.
Therefore, a method that can enable server nodes to share memory resources easily is needed.